[Bug]: RAPTOR and

[alexff77]

Is there an existing issue for the same bug?

• I have checked the existing issues.

RAGFlow workspace code commit ID

9298acc full - Nighty Feb 18

RAGFlow image version

9298acc full

Other environment information

Running nightly build from Feb 18 on ubuntu

Actual behavior

Hi,
I have several issues similar to what others reported but not quite the same.
Issues with RAPTOR:
1) On this one doc with only 1 chunk it always errors our:

15:10:00 Task has been received.
15:10:01 Page(12): OCR started
15:10:04 Page(12): OCR finished (3.11s)
15:10:05 Page(12): Layout analysis (0.86s)
15:10:05 Page(12): Table analysis (0.00s)
15:10:05 Page(12): Text merged (0.06s)
15:10:05 Page(12): Page 01: Text merging finished
15:10:05 Page(12): Generate 1 chunks
15:10:05 Page(12): Embedding chunks (0.22s)
15:10:05 Page(12): Done (0.04s)
15:10:08 Start RAPTOR (Recursive Abstractive Processing for Tree-Organized Retrieval).
15:10:08 Task has been received.
15:10:08 [ERROR]Fail to bind LLM used by RAPTOR: 'NoneType' object is not subscriptable
15:10:08 [ERROR][Exception]: 'NoneType' object is not subscriptable

I can find chunk in elasticsearch.

2) On another document it processed fine,
then I changed a setting on the file to enable Entity resolution and re-run it and got an error.

16:01:12 Reused previous task's chunks.
16:01:17 Start RAPTOR (Recursive Abstractive Processing for Tree-Organized Retrieval).
16:01:17 Task has been received.
16:01:18 [ERROR]Fail to bind LLM used by RAPTOR: 'NoneType' object is not subscriptable
16:01:18 [ERROR][Exception]: 'NoneType' object is not subscriptable

Then I turned that setting off and re-run and RAPTOR worked fine again. This time it had to re-generate chunks since it's error-ed before.
Simmilar for another document I chose not to regenerate chunks and it failed RAPTOR, they I re-generated and RAPTOR worked.

I see that I have some errors connecting to the elasticsearch

ESConnection.update got exception: BadRequestError(400, 'illegal_argument_exception', 'exceeded max allowed inline script size in bytes [65535] with size [213572] for script [ctx._source.content_with_weight='

Before I had errors regarding number of scripts that can be run and I increased it to 1000/1m

Could be related to how many entities it found and trying to resolve?

3) Tasks seems to be stuck at very last step after entity resolution for a really long time, for an hour or more for example.
18:57:39 Entities extraction progress ... 46/47 (8962 tokens)
18:57:39 Entities extraction progress ... 47/47 (9589 tokens)
then some times it fails.

Thank you.

Expected behavior

No response

Steps to reproduce

As described above. Using documents with RAPTOR and entity resolution.

Additional information

No response

[KevinHuSh]

About 15:10:08 [ERROR][Exception]: 'NoneType' object is not subscriptable, do you have back end error log?
docker logs -f ragflow-server

For illegal_argument_exception:

curl -X POST  -u elastic:infini_rag_flow -H 'Content-Type: application/json'  http://127.0.0.1:1200/_cluster/settings -d '{"transient":{"script.max_size_in_bytes": 10000000}}'

[alexff77]

1. For 15:10:08 [ERROR][Exception]: 'NoneType' object is not subscriptable

functools.partial(<function set_progress at 0x794cf594b7f0>, '47c8a1b2f08711ef817e0242ac120007', 100000000, 100000000)

` }

if row["pagerank"]:

    doc[PAGERANK_FLD] = int(row["pagerank"])

res = []

tk_count = 0

for content, vctr in chunks[original_length:]:

    d = copy.deepcopy(doc)

    d["id"] = xxhash.xxh64((content + str(d["doc_id"])).encode("utf-8")).hexdigest()

    d["create_time"] = str(datetime.now()).replace("T", " ")[:19]

    d["create_timestamp_flt"] = datetime.now().timestamp()

    d[vctr_nm] = vctr.tolist()`

2. For 'NoneType' object has no attribute 'strip'

handle_task got exception for task {"id": "ff21e0a6f16f11efa8db0242ac120007", "doc_id": "ca0f4d84ebe511ef801f0242ac120007", "from_page": 0, "to_page": 100000000, "retry_count": 0, "kb_id": "283fb782ebe511efb7e50242ac120007", "parser_id": "naive", "parser_config": {"auto_keywords": 2, "auto_questions": 0, "raptor": {"use_raptor": true, "prompt": "Please summarize the following paragraphs. Be careful with the numbers, do not make things up. Paragraphs as following:\n {cluster_content}\nThe above is the content you need to summarize.", "max_token": 256, "threshold": 0.1, "max_cluster": 64, "random_seed": 0}, "graphrag": {"use_graphrag": true, "entity_types": ["organization", "person", "geo", "event", "category", "procedure", "drug", "sample", "biological sample", "organ", "chemical", "document", "process", "regulatory body", "consumables", "medication", "equipment"], "method": "light", "resolution": true}, "chunk_token_num": 128, "delimiter": "\n!?;\u3002\uff1b\uff01\uff1f", "pages": []}, "name": "Secti...

`rag/llm/chat_model.py in chat at line 50
In App

    try:

        response = self.client.chat.completions.create(

            model=self.model_name,

            messages=history,

            **gen_conf)

        ans = response.choices[0].message.content.strip()

        if response.choices[0].finish_reason == "length":

            if is_chinese(ans):

                ans += LENGTH_NOTIFICATION_CN

            else:

                ans += LENGTH_NOTIFICATION_EN`

** 3. Increasing script size helped with this error, thank you, but now I'm getting another error**

`ESConnection.update got exception: BadRequestError(400, 'script_exception', 'compile error')ctx._source.content_with_weight='{ "directed": false, "multigraph": false, "graph": {}, "nodes": [ { "entity_type": "ORGANIZATION", "rank": 56, "pagerank": 0.007729082736942685, "id": ....

I tried to increase size even more, not sure if that's the solution.

[alexff77]

Increasing script size didn't help, still getting
ESConnection.update got exception: BadRequestError(400, 'script_exception', 'compile error')

Thanks.

[alexff77]

Latest update seem to fixed most of these issues.
Still have issues with really long processing of documents.
Tasks seems to be stuck at the last step